Topics
Artificial Intelligence and Business Strategy
In collaboration with
BCG
In collaboration with
Why does how you describe your team — down to its name — matter? Gerri Martin-Flickinger, former executive vice president CTO at Starbucks, joins the Me, Myself, and AI podcast to describe some of the technology initiatives the coffeehouse chain has been able to pursue since rebranding its technology team and articulating its mission.
In her conversation with hosts Sam Ransbotham and Shervin Khodabandeh, Gerri recaps a decades-spanning career working in technology leadership roles at Chevron, McAfee, and Adobe, then explains some recent employee- and customer-facing projects Starbucks has undertaken using AI and machine learning.
Read more about our show and follow along with the series at https://sloanreview-mit-edu.ezproxy.canberra.edu.au/aipodcast.
Subscribe to Me, Myself, and AI on Apple Podcasts, Spotify, or Google Podcasts.
Transcript
Sam Ransbotham: What’s in a name? Today we talk with Gerri Martin-Flickinger, former chief technology officer at Starbucks, about how the names we use can make a big difference in innovation and motivation.
Welcome to Me, Myself, and AI, a podcast on artificial intelligence in business. Each episode, we introduce you to someone innovating with AI. I’m Sam Ransbotham, professor of information systems at Boston College. I’m also the guest editor for the AI and Business Strategy Big Ideas program at MIT Sloan Management Review.
Shervin Khodabandeh: And I’m Shervin Khodabandeh, senior partner with BCG, and I colead BCG’s AI practice in North America. Together, MIT SMR and BCG have been researching AI for five years, interviewing hundreds of practitioners and surveying thousands of companies on what it takes to build and to deploy and scale AI capabilities and really transform the way organizations operate.
Sam Ransbotham: Today we’re talking with Gerri Martin-Flickinger, former executive vice president and chief technology officer at Starbucks. Gerri, thanks for taking the time to talk with us. Welcome.
Shervin Khodabandeh: It’s really great to have you here, Gerri.
Gerri Martin-Flickinger: It’s great to be here. Thanks for having me.
Shervin Khodabandeh: So, Gerri, tell us a little bit about yourself — your background, your journey to this point, and what it’s been like.
Gerri Martin-Flickinger: This topic today is going to be about AI, so I would really love to go way back in time to the 1980s. I know you probably didn’t expect me to start there when you asked me that question, but I have to start there, because I went to Washington State University — go Cougs! — and I had an emphasis in artificial intelligence way back in the day and, in fact, did my senior project with a neural net that was built in Lisp, which … nobody even remembers what that is anymore. But the reason why I start there is because I have had a love for AI and the potential of AI techniques for all of this time. And it’s been really exciting in the last five or six years to see that we’re finally at a place where technology, both compute and storage — and data — have gotten to the point where we can actually start to achieve some of those visions we had way back then.
But let me start in ’85. After going to school and getting a degree in computer science, I went to Chevron, the oil/energy company. [I] spent my early years there and did some really cool AI work while I was there, which was really more in a research mode, and did some technology that actually was sold and purchased by some other companies outside of Chevron.
After that, I became the first CIO for McAfee in the late ’90s, when McAfee was still quite small but grew quite large. And that was really fun, because if you think about antivirus in those early days, it was one of the first SaaS companies, because you would buy your antivirus software, but then you would get these payloads every month that would be the new virus signatures to keep protecting your machine. So, in effect, it was like a subscription business. Being part of that in the early days of, frankly, the internet and right during the whole dot-com period was super insightful and really taught me a lot about scale and consumer digital before we called it consumer digital.
I then took a little bit of time out of my career to have two beautiful twin daughters, and then I came back as the CIO for Verisign, another security company, and then became the CIO for Adobe and was at Adobe for about 10 years and part of the team who migrated their product offerings into the cloud, into a true subscription SaaS business.
And then [I] joined Starbucks six years ago, a little over six years ago, as their first chief technology officer and really helped them migrate and move into a more modern architecture stack; evolve their entire digital platform, including mobile order and pay; [and] innovate IoT into all of the stores — so really, just a whole lot of really fun at-scale technology and, in the process, got back to my roots and did some AI at the same time.
Shervin Khodabandeh: This is a phenomenal series of things you’ve been part of. And I’m curious, as someone whose interest and exposure and practicing of AI goes back to the ’80s, as you said, and you’ve seen the various eras of technological and organizational innovations, and the playing field has continuously increased — what do you think are some of the biggest misconceptions that still exist in the minds of executives, particularly when it comes to this topic?
Gerri Martin-Flickinger: That is a great question. One of them is that if you build a model, it can solve anything. There’s no such thing as a generalized AI model that will solve anything or everything. That just isn’t possible. If you think about where people are having the most success right now with AI at scale, a lot of it is tightly coupled to statistical analysis, frankly, and is a lot about taking very large learning sets and building very sophisticated models that can be predictive in nature. And that’s awesome, and what that takes is a lot of data to make it accurate.
And so I think one of the misconceptions that I’ve certainly talked to a lot of executives about in the last few years is that in order to do AI in a really meaningful way, you’ve got to get your data in order. It isn’t as simple as just saying, “Hey, we have loads of data! We should be able to have amazing AI models.” You’ve got to have a little bit of structure to that data. You have to have a little bit of thought about where that data sits, even something as simple as, what does your data lake look like? Where are you putting that data? What is the currency of that data? Do you need the models to retrain in real time? Do you need to build them once and then retrain them once a quarter? And all of that starts to get really wrapped up into your data architecture. So I think the one thing that I always ask people when they want to talk about AI is, “Tell me about your data. Do you have data? Where is your data? Do you own it? Can you use it? Do you have the right to use it?”
This is the other thing about consumer data that’s so very important: You have to make sure you’re not doing things that you shouldn’t be doing with that data. So I really think that one of the misconceptions out there is this idea that there’s this thing you buy called AI, and you plug it in and it all works. That is almost the end of a long series of things you need to do and architect.
Sam Ransbotham: OK, so, once an organization has their data in place, what happens next? What does having that data enable?
Gerri Martin-Flickinger: I do think the evolution of AI and ML specifically has led all parts of businesses that are sophisticated to ask very different questions than they did even five years ago. Suddenly, there’s an expectation of, “No, we should be able to predict supply chain forecast based on X, Y, and Z.” And that isn’t going to take somebody to build a giant spreadsheet to do it. There’s a better way to do that now. And I think you’re seeing that evolution. What I think I would compare this to, though, is … do we all remember the days before we all had Excel? It’s hard to remember those days, but —
Sam Ransbotham: VisiCalc?
Gerri Martin-Flickinger: Yeah, VisiCalc! I remember VisiCalc. But there was a time —
Shervin Khodabandeh: Lotus 1-2-3.
Gerri Martin-Flickinger: There was a time when the idea that all of us in business would be able to build a spreadsheet and then share it and collaborate on it. That was like, “What? Why would you …”
Sam Ransbotham: Crazy talk.
Gerri Martin-Flickinger: Crazy talk, right? Same thing when I first started working; there was a typing pool. We didn’t have email; you sent memos. My point being that as tools become more available to more people, more people are able to explore more ideas. How many times do all of us open spreadsheets all week long to do everything from help our kids with homework to track our personal finances, or just keep a list? The best list thing I have is a spreadsheet.
And suddenly, that has changed how we all work. So I just go back to these fundamental shifts we’ve seen in the past, like the spreadsheet, and I don’t think this is so very different. I think as we continue to watch this evolve, we’re going to find better tools and more effective ways that we can all explore these techniques.
I’ll use a couple of examples from Starbucks: doing models for labor scheduling. That’s not a leap to think about; certainly, [it’s] something that you can do with the kind of data that you have today.
Now, do you think a store manager knows or cares that when they build a labor schedule for their store, there’s actually a reasonably sophisticated ML model behind the scenes doing that? No. So I think the amount of embedded AI and ML that we all … We all probably have some in our cars right now; a lot, probably. Your power companies, your phone companies … It’s embedded everywhere. Your credit card companies have had it probably longer than you even realize, for fraud detection, so it’s already pretty pervasive. And I think the question is, how much more accessible could it be as people become more sophisticated? I don’t know.
I have kids in high school. Kids in high school talk about data science now; it’s classes in high school, so there’s nothing to think that in five, six, seven years, when they come out of college, they’re going to be probably pretty fluent in some of these techniques, even if it seems completely inconceivable to all of us.
Shervin Khodabandeh: No, I think … it actually reminds me of a Wall Street [Journal] article I was reading today about chess. Chess grandmasters have had to become experts, somewhat, in AI, because everybody’s using it. And they have these AI teams to be able to handicap different lines of thinking of the algorithm to throw off their opponents, because everybody’s using AI to plan and win their games. And so you need to really understand how the engine works if you’re going to beat somebody who’s using that engine to beat you. And it builds on the point you were making on, it is quite pervasive.
Sam Ransbotham: In your analogy, I guess, that would be “Out-schedule the competition.” You mentioned Starbucks. Is there something particular that you’re excited about that you want to showcase?
Gerri Martin-Flickinger: I can certainly highlight a couple of examples. We have a moniker, Deep Brew, which stands for a broad section of AI projects underway across the company. The ones that folks are most familiar with are personalization models.
Whether it’s personalization on the mobile app or personalization when you pull into a drive-through where they have a digital display, those experiences are being driven from models that are based on lots of different inputs, some of which are very personal, like maybe your own buying patterns. Some of them are regional, like, “What’s going on with buying patterns in this region?” They could be environmental factors, like, “What’s the weather today?” It could have to do with supply chain loads, like, “What do we actually have in stock that we need to sell?”
Those are actually much harder to do than they sound. It sounds very simple. But let me give you an example to illustrate why some of this is hard and why I always start with data. Doesn’t it sound easy to figure out if there are the ingredients for a latte so that you can promote a latte on the phone?
Sam Ransbotham: Naively, I’ll say yes.
Gerri Martin-Flickinger: It sounds like it’s a thing, right? Like, “Yeah, we have lattes.” Well, actually, lattes are manufactured in the moment at a store, and they’re made of component parts. They’re made of some espresso, which could be different kinds of espresso, made with some type of a milk product, which could be cow milk, it could be an alt [nondairy] milk. And then it could be heated to different temperatures based on what the customer has asked for. And that’s a simple drink. That is the simplest espresso drink, probably, you can get in a store.
So here’s why that’s complicated. If you’re in the store, the customer just knows it as a latte. But if you think about the entire supply chain of all of the component parts that have to be available at that moment in the front of the house, behind the counter, to make that latte, that’s a whole different problem. And now you’ve got to get all the way back to your data master, the data master that is the component parts, and understanding if they were delivered that night in the back of the store.
Now you’ve got a data problem that requires you to decompose and restructure the data all the way back to the origin, if you haven’t already done that. And I’m only illustrating this because so often you think, “Well, it’s an easy ML problem to say, ‘We want to promote lattes.’” But the second you do that, you actually have to know the deepest level of data possible to ensure you actually have the product to sell.
Sam Ransbotham: That’s tricky because, actually, when you say “latte,” I know exactly what you mean, because you mean exactly the one that I would drink. You don’t mean the one that Shervin would drink.
Gerri Martin-Flickinger: Right.
Sam Ransbotham: And to answer the question, you’ve got to answer it for every single person.
Gerri Martin-Flickinger: Right. And you’re not going to enumerate all those. There’s infinite possibilities for customization. Infinite. So you can’t do that. You have to actually work at it as a data problem. And then you can do the AI model on top of it, because you’ve actually figured out what you have to work with.
Shervin Khodabandeh: And then you also talked about the supply chain issue and the inventory management. And the point to me is, these use cases are not in silos anymore.
Gerri Martin-Flickinger: Exactly.
Shervin Khodabandeh: The whole foundational data, of course, is critical to power them. But how we market impacts what happens in the store, and supply chain issues impact what we should be able to market or shouldn’t. And so —
Gerri Martin-Flickinger: Totally.
Shervin Khodabandeh: My follow-up is, for the business leaders who are listening to this, I think there are many analogs of what you just described that would resonate in any line of business, because you’ve got these groups that are different lines of business or different functional components that, in today’s world with today’s data, are much more interactive, and there’s a network effect of all of these things, which requires teams to come together that normally wouldn’t work together.
What advice do you have for the CEO or the president of a business unit to break these silos, because you’ve got different teams with different tools, different incentives, right? That must be a daunting organizational problem. It’s not just a technology problem. And you’ve seen that work well. I’m just curious — what advice would you have?
Gerri Martin-Flickinger: Well, I don’t think there’s any magic here. I think, as in most things in business, you have to start by being really clear on, what is the objective? What are you solving for? That sounds so simple, but sometimes that’s really hard to figure out. Is what you’re solving for increased revenue? Is it increased customer retention? Is it improved margin? Is it something else?
Getting really clear on that is part of what gets all the constituencies to go at loggerheads — somebody carries the hat of revenue, and somebody else carries the hat of margin, and somebody else carries customer experience. You’ve got to get clear on what you’re solving for. And you can’t solve for all of it at the same time. Now, you can benefit it all, but you have to get really clear on “What are we going after?” So that’s my first advice: [to] be really clear on what problem you’re trying to solve.
I think one thing … I believe a lot in bringing people together who have different expertise. I actually think it’s a good thing. It’s a good thing to bring people together who have five different specialties, because they’re going to bring the very best thinking for that domain. But then you also have to have them feel like they’re in it together, and that’s good, old-fashioned teamwork. And I hate to say “good, old-fashioned teamwork,” but for as long as I’ve been in the business world, it all comes down to the same things: Are you getting people together with a common vision? Are you giving them room to fail so that they can get onto a path to success? Are you giving them a goal that’s really clear, with a timeline that’s achievable but also really clear? And then are you supporting them with the resources and the budget that they need to be successful?
It’s all that same stuff. There’s nothing new there. I do think where a lot of people fail is, they don’t start off with a clear problem they’re trying to solve. And that tends to get people to all get really entrenched in their silos and then go off and try to solve their own problem.
Shervin Khodabandeh: That’s very well said.
Sam Ransbotham: I’m struck as I’m listening about how much depth you obviously have in making a latte, but that wasn’t your background; we didn’t hear that stop [in your career journey]. How do you get people to know so much about the domain area to then be able to solve it with the technology that you’re using to solve it? It … seems like a very difficult thing to pull together?
Gerri Martin-Flickinger: I don’t know which question to answer: the one about how did I end up learning to make a latte at Starbucks or, in general, how do you do that in business?
Shervin Khodabandeh: Let’s start there.
Gerri Martin-Flickinger: OK, we can start with my journey. So, yeah, I came out of technology. I had been in enterprise software for many, many years in Silicon Valley. And my decision to come to Starbucks was kind of interesting. I was, first of all, just intrigued. I was intrigued by the scale. And the scale is interesting when you think about Starbucks, because today there’s over 30,000 stores around the world. There’s over 300,000 baristas around the world.
And why it’s an interesting scale problem is not just the number of customers that visit Starbucks. But if you think about those 30,000-some stores, each one is like a little business unto itself. When you’re in enterprise software, you might have a hundred offices around the world. You might have a ton of people, but they’re in these big offices with big pipes and lots of infrastructure, and you have a support team there.
When you’ve got a store in the middle of Oklahoma on a dial-up line, that’s a whole different thing to manage. And to have the same expectation of quality for a customer who’s got their mobile order-and-pay application, that’s just a whole different game. And I was really intrigued by, of course, IoT and how much more could be done in brick-and-mortar [retail] with IoT devices. I was intrigued by how quickly I saw consumer digital growing. And so all those things are what made me come to Starbucks and be part of that transformation.
How do you learn when you know nothing about food and beverage? First thing you do is, you spend time in the stores, you know? I spent my first few weeks in a store learning about how people make lattes. Now, I cannot claim to be a barista by any stretch at all, so when I was in the store, I was mostly helping clean, or I was greeting customers, or I was trying to do things I could actually do. But in the process, you learn a lot about what goes on in a store — and not just the really cool stuff that you see, like making the lattes or greeting the customers, but what goes on in the back of house. How do they receive inventory? Oh my gosh. How do they do payroll? How do they have to do labor scheduling? What does that look like? And that’s an eye-opener.
And I would say, forget what business you’re in. Whatever business you’re in, if you’re a technologist, if you are not sitting shoulder [to] shoulder with whoever is the tip of the spear of the business, you’re missing an opportunity. You’ve got to do that. So when I was in enterprise software, I spent a lot of time going out with salespeople to visit customers. I just wanted to see, what are the customers thinking? Do they love us? Do they hate us? What problems are they having? Spending time with the customer support center to just sit down and listen, occasionally, to the calls they were getting: What is the world thinking about us, and how are people who are depending on our software feeling?
Shervin Khodabandeh: Did you spend time on the oil rig when you were at Chevron?
Gerri Martin-Flickinger: I did spend time at refineries. Yeah, a little bit of time at refineries.
Shervin Khodabandeh: That’s great.
Sam Ransbotham: Actually, it reminds me of Prakhar Mehrotra and Walmart. He went out there and — this was one of our earlier interviews — said a very similar thing about how you understand how to automate or how to put technology into these situations. And it was very much echoing the kinds of things you’re saying — [it] can’t be done in isolation.
Gerri Martin-Flickinger: That’s right. And I do think there’s a couple of things that, in my playbook, have continued to pay off over and over again. And they’re just simple, simple things. The first is, words matter. Words matter. They really do. When I say to you — and I’m going to ask both of you to answer back at me; I’m going to ask you a question now: When I say the word IT, what do you think of?
Shervin Khodabandeh: I think of email servers and software patches and things like that.
Gerri Martin-Flickinger: Sam, what about you?
Sam Ransbotham: I think I’m biased. I thought more of a strategy-oriented, how you’re enabling connectivity within the organization —
Gerri Martin-Flickinger: We’re not using your answer. No, no, no.
Sam Ransbotham: OK. I’m an IT prof, so that’s maybe my bias there. Do you think most people go, “Operations”?
Gerri Martin-Flickinger: When I say IT, most people will talk about the help desk; they’ll talk about outages.
Sam Ransbotham: Trouble tickets.
Gerri Martin-Flickinger: They’ll talk about services, trouble tickets, data centers. Right? OK. When I say technology, people say, “The future. Innovation. New things.” Right? So if you’re in a business, and someone introduces someone who’s in the IT department, they have one reaction. If I introduce someone to you and I say, “This is” — in the case of Starbucks — “Starbucks Technology,” which one sounds and feels more future-leaning?
Sam Ransbotham: Actually, that’s a huge difference.
Gerri Martin-Flickinger: It’s a huge difference.
Sam Ransbotham: I definitely see the difference there.
Gerri Martin-Flickinger: Right. Which is why I said words really matter. And so we were talking about transformation and how do you transform a technology organization to the future. And so, one thing that’s one of these tried-and-true things is, did you name your organization in a way that makes the organization proud, so that every single person in the organization sits up a little straighter and maybe works a little harder? Have you named the organization in a way that really represents what you want it to become? And have you named it in a way that everyone else in the business looks at it and goes, “Oh, that’s something a little different”?
OK. I know a long answer to one of the things that I think is really important in transformation is to signal that you’re doing it. And so, for example, with Starbucks, when I joined, 90 days after I joined, [I] changed the name of IT to Starbucks Technology. Never used the word IT again.
And if I was ever in a meeting where somebody said IT, I’d stop the meeting and I’d say, “We don’t have IT. We have Starbucks Technology,” and it’s kind of funny, because that one change made a big difference.
The next thing that I think can make a difference is, you need a tagline. I hate to say it, but everybody in business, everyone who’s a CEO, knows it. You’ve got to tell your story, and you don’t get five hours to tell your story. You get six to 10 words, and you’d better get people curious to ask more. And so, put a tagline in place. Super simple: “Talented technologists delivering today, leading into the future. Starbucks Technology.” That’s it.
And that simple phrase, which is used today still, after six years, just continues to reinforce the value of the organization, the value of the people, the importance of getting the work done — as well as continuing to build for the future. And so, for me, transformation comes down to people. And to do any transformation with tech has nothing to do with the tech as much as it has to do with the people who are making it happen. They have to feel inspired, they have to feel what they’re doing is important, and they have to feel like they have room to be part of the invention of the future. And I think that’s all we have to do as leaders, is make room for that.
Sam Ransbotham: Gerri, it was great talking with you. [You have] such a vast experience and a great ability to connect those experiences together to give us a holistic view of what’s happening and what may happen in the future. Thank you for taking the time.
Shervin Khodabandeh: It’s been really wonderful. Thank you.
Gerri Martin-Flickinger: It’s been fun. Thanks so much.
Sam Ransbotham: Next time, Shervin and I talk with Barbara Martin Coppola, the chief digital officer for IKEA Retail. Join us as we hear what Barbara thinks about the meaning behind the words we use when we talk about artificial intelligence.
Allison Ryder: Thanks for listening to Me, Myself, and AI. We believe, like you, that the conversation about AI implementation doesn’t start and stop with this podcast. That’s why we’ve created a group on LinkedIn, specifically for leaders like you. It’s called AI for Leaders, and if you join us, you can chat with show creators and hosts, ask your own questions, share your insights, and gain access to valuable resources about AI implementation from MIT SMR and BCG. You can access it by visiting mitsmr.com/AIforLeaders. We’ll put that link in the show notes, and we hope to see you there.
